ISBM College of Engineering
Pune, Maharashtra
Savitribai Phule Pune University
Affiliated University • A.Y. 2026–2027
AI-Powered Smart Marketplace & Procurement Platform
Intelligent Multi-Marketplace Search, Comparison & Recommendation Engine
Bachelor of Engineering (B.E.) • Department of Artificial Intelligence & Data Science
4
Members
8
Months
6+
AI Models
5
Marketplaces
Team — Failed Engineers United Club Pvt. Ltd.
Snehal Dhumal Sahil Dhumal Yash Warulkar Shubham Ghodke
🎓 Project Guide — Prof. Anil Walke
Failed Engineers United Club Pvt. Ltd.
01 / 18
Overview

Presentation Agenda

What we'll cover today — sixteen sections outlining vision, design, and delivery.

01
Problem Statement
Why current solutions fail
02
Existing Solutions
Market analysis & gaps
03
Proposed Solution
AI marketplace architecture
04
System Architecture
Pipeline & module design
05
Novelty & Research Gap
What makes us unique
06
Project Objectives
Strategic goals
07
Core Methodology
AI approach & RAG
08
Technology Stack
Tools & infrastructure
09
Development Roadmap
4-phase lifecycle
10
Detailed Timeline
Week-by-week plan
11
Budget Breakdown
Cost analysis
12
Risk & Mitigation
Risk management
13
Evaluation Metrics
Benchmarks & KPIs
14
Future Scope
Expansion avenues
15
Summary Timeline
8-month overview
16
Conclusion
Thank You & Q&A
Failed Engineers United Club Pvt. Ltd.
02 / 18
Challenge

The Problem

Why current marketplace & procurement systems are broken for consumers and businesses.

75%
Time spent on
manual comparison
5+
Platforms visited
per purchase
40%
Procurement teams
overspend
3hrs
Average research
per decision
👤
Individual Users – Consumers
  • ▸ Visit multiple websites to compare products and prices
  • ▸ Difficulty identifying best option by budget & quality
  • ▸ Time-consuming decision-making with no AI help
  • ▸ No cross-platform intelligent recommendations
🏢
Businesses – Procurement Teams
  • ▸ Manual research to find reliable suppliers
  • ▸ Hard to compare quotations, delivery & credibility
  • ▸ Slow and inefficient procurement cycles
  • ▸ No AI-driven vendor evaluation pipeline
Multiple Sites Fragmented search Manual Compare Hours of effort Decision Fatigue Information overload Sub-optimal Wrong choices made ✗
🎯 Core Insight No single AI platform understands user needs and recommends options across all marketplaces — making comparison slow and inaccessible.
Failed Engineers United Club Pvt. Ltd.
03 / 18
Landscape

Existing Solutions & Their Limitations

Market analysis — none combine AI-driven search with multi-marketplace aggregation.

PlatformPrimary PurposeKey LimitationPrice Barrier
AmazonProduct ShoppingLimited to its own marketplaceNo cross-platform
FlipkartOnline ShoppingDoes not compare competitorsSingle platform
IndiaMARTB2B ProcurementRequires manual supplier evaluationPremium paid
Fiverr / UpworkService HiringFocused only on freelancersCommission-based
RentomojoProduct RentalsLimited to rental servicesSubscription
Google ShoppingPrice ComparisonNo AI recommendationsAd-driven results

🔬 Critical Gap Identified

No single platform combines:

  • AI-driven natural language search & understanding
  • Multi-marketplace aggregation across products, services & rentals
  • Unified comparison with intelligent ranking
  • One-platform recommendations at affordable price point
Failed Engineers United Club Pvt. Ltd.
04 / 18
Solution

Proposed Solution

AI-powered marketplace platform with intelligent search, comparison & recommendation engine.

🗣️ User Natural Language 🤖 NLU Intent Parsing 🔍 Search Multi-Platform ⚖️ Compare Price · Quality 📊 Rank AI Scoring ✓ Best Pick
🗣️
Natural Language Search
Ask in plain English — AI understands intent
⚖️
AI Comparison Engine
Compare price, quality, delivery across platforms
🧠
Smart Recommendations
Context-aware, budget-optimized suggestions
⏱️
Time & Cost Savings
One query replaces hours of manual research
Failed Engineers United Club Pvt. Ltd.
05 / 18
Architecture

System Architecture

A modular pipeline transforms a natural-language request into ranked, contextual recommendations.

👤 User Query Input FRONTEND Next.js + React Tailwind CSS UI API GATEWAY FastAPI REST + WebSocket AI ENGINE LangGraph Agent Orchestration NLU → Search → Compare → Rank → Recommend 🗄️ Qdrant Vector Embeddings LLM MODELS Llama · Qwen Understanding & Comparison 🛒 Amazon Product API 🛍️ Flipkart Product API 🏭 IndiaMART B2B API 💼 Fiverr Services API 🏠 Rentomojo Rental API 🗃️ PostgreSQL MARKETPLACE ADAPTER LAYER — UNIFIED API INTERFACE Each marketplace adapter is isolated — API changes don't affect the core engine
Failed Engineers United Club Pvt. Ltd.
06 / 18
Innovation

Novelty & Research Gap

What existing systems lack vs. what Zentro delivers — a comprehensive feature comparison.

Feature / CapabilityExisting SystemsZentro (Ours)
Single Marketplace Search✓ Yes✓ Yes
Basic Product Listings✓ Yes✓ Yes
Multi-Marketplace Aggregation✗ No✓ Yes
Natural Language Search (NLU)✗ No✓ Yes
AI-Powered Comparison Engine✗ No✓ Yes
Intelligent Ranking & Recommendations✗ No✓ Yes
Unified Products + Services + Rentals✗ No✓ Yes
RAG-based Context-Aware Results✗ No✓ Yes
Supplier Credibility Scoring✗ No✓ Yes
🚀 Key Innovation Zentro is the FIRST platform to combine AI-powered natural language search with multi-marketplace aggregation, intelligent ranking, and unified comparison — all in a single, affordable platform.
Failed Engineers United Club Pvt. Ltd.
07 / 18
Goals

Project Objectives

Eight strategic goals driving the development and impact of Zentro.

Failed Engineers United Club Pvt. Ltd.
08 / 18
Approach

Core Methodology

Building a hybrid AI pipeline with RAG-based retrieval and LLM-powered intelligence.

ApproachQualityFeasibilityAcademic Value
Direct LLM API callsGood but expensive✓ Yes★☆☆ Low
RAG + Vector SearchHigh with context✓ Yes★★☆ Medium
LangGraph Agent WorkflowBest orchestration✓ Yes★★★ High
CHOSEN: RAG + LangGraph + Multi-LLMBest overall✓ Yes★★★ Highest
DATA SOURCES 📦 Marketplace APIs EMBED 🔢 Text Embeddings 🗄️ Qdrant Vector DB RETRIEVE 🗣️ User Query AUGMENT 📝 Context + Query GENERATE 🧠 LLM (Llama/Qwen) ✓ RAG = Retrieval-Augmented Generation — context from real marketplace data enriches every LLM response
Failed Engineers United Club Pvt. Ltd.
09 / 18
Stack

Technology Stack

A modern, production-ready toolchain — 100% free & open source.

FRONTEND Next.js React · Tailwind BACKEND FastAPI Python 3.12 DATABASE PostgreSQL Relational Data AI FRAMEWORK LangGraph LangChain VECTOR DB Qdrant Embeddings AI MODELS Llama · Qwen Open Source
Deployment
Docker · Vercel · Render
💡 Budget-Friendly All software & AI models are 100% free/open-source. Total cost: ₹0 – ₹5,000 for 8 months.
Failed Engineers United Club Pvt. Ltd.
10 / 18
Timeline

Development Roadmap

A four-phase development lifecycle from discovery to launch.

Phase 1 — Discovery
  • Requirement Analysis
  • System Design & Architecture
  • UI/UX Wireframes (Figma)
  • Data Collection & APIs
    • Phase 2 — Build
  • Backend Development (FastAPI)
  • AI Pipeline Integration
  • Database Schema Design
  • RAG Pipeline Setup
    • Phase 3 — Integrate
  • Marketplace API Integration
  • Recommendation Engine
  • Comparison & Ranking
  • End-to-End Testing
    • Phase 4 — Launch
  • Docker Deployment
  • Beta Testing (20 Users)
  • Performance Optimization
  • Final Submission & Demo
    • Failed Engineers United Club Pvt. Ltd.
    11 / 18
    Execution

    Detailed Week-by-Week Timeline

    All four phases — Foundation through Polish & Deployment.

    PHASE 1 Research & Foundation (Weeks 1–4)
  • W1 Project Setup: GitHub monorepo, dev environment, scope document
  • W2 Architecture Design: System diagram, DB schema, API contracts
  • W3 Data Collection: Marketplace APIs research, scraping prototypes
  • W4 Basic Prototype: Auth, search interface, API connectivity
    • PHASE 2 AI Pipeline Build (Weeks 5–12)
  • W5–6 LangGraph agent setup, NLU query processing pipeline
  • W7–8 Vector database (Qdrant) setup, embedding pipeline
  • W9–10 RAG pipeline: retrieve marketplace data, augment queries
  • W11–12 Multi-LLM comparison (Llama vs Qwen), production model selection
    • PHASE 3 Platform Intelligence (Weeks 13–20)
  • W13–14 Comparison engine: weighted scoring across price, quality
  • W15–16 Recommendation engine: personalized rankings, budget awareness
  • W17–18 Dashboard & analytics: results visualization, export
  • W19–20 Marketplace integration: real-time pricing, availability
    • PHASE 4 Polish & Deployment (Weeks 21–32)
  • W21–24 UI/UX overhaul, beta testing with 20 users, bug fixes
  • W25–28 Performance optimization, security hardening, Docker deploy
  • W29–30 Thesis writing: 7 chapters covering architecture & results
  • W31–32 Demo preparation, final submission, team post-mortem
    • Failed Engineers United Club Pvt. Ltd.
    12 / 18
    Budget

    Budget Breakdown

    Zero-cost software stack — only hosting & domain expenses for production.

    ₹0
    Development Software
    Open source / free
    ₹0
    AI Models (Llama, Qwen)
    HuggingFace / Ollama
    ₹0
    Frontend (Vercel)
    Free tier
    ₹0
    Backend (Render)
    Free tier
    ₹0
    Database (Supabase)
    Free up to 500 MB
    ₹0
    Vector DB (Qdrant)
    Self-hosted
    ₹500
    Domain Name
    Optional
    ~₹1.5K
    Compute & Electricity
    API + GPU
    ₹2,000 – ₹5,000
    TOTAL (8 months, 4-member team)
    ✓ 100% free/open-source tools ✓ Affordable for any institution ✓ Commercial: ₹50,000+/year
    Failed Engineers United Club Pvt. Ltd.
    13 / 18
    Risk

    Risk Register & Mitigation

    Proactive risk management for project success.

    RiskLikelihoodImpactMitigation Strategy
    API rate limiting / blockingHighMediumCaching, proxy rotation, fallback data
    LLM response quality too lowMediumHighRAG + prompt engineering + multi-model fallback
    Marketplace API changesMediumMediumAdapter pattern — isolate marketplace logic
    Scope creepHighHighLock scope after Month 1; new ideas → v2 backlog
    Team member unavailabilityMediumHighDistribute knowledge; no single point of failure
    LLM latency >5sMediumMediumStream tokens; Groq API free tier for demo
    Live demo failureLowVery HighPre-record 5-min backup video
    ⚠️ Risk Heat Map Red = High Priority  |  Amber = Monitor  |  Green = Low (backup ready)
    Failed Engineers United Club Pvt. Ltd.
    14 / 18
    Metrics

    Evaluation Metrics

    Quantitative performance benchmarks and KPIs.

    MetricWhat It MeasuresTool / MethodTarget
    Search RelevanceDoes result match intent?LLM-as-judge prompt> 4.0/5
    Comparison AccuracyAre rankings objectively better?Human rubric (3 reviewers)> 3.5/5
    Response LatencyQuery to recommendationsPython time (50 samples)< 5s
    User SatisfactionPlatform usefulnessGoogle Form (20 users)> 4.0/5
    Cost Savings% vs manual comparisonTime tracking> 60%
    Platform CoverageMarketplaces searchedAPI integration count5+
    RAG + LangGraph Pipeline
    Academic Value: ★★★
    Production Quality: ★★★★
    Latency: ★★★★
    Direct API (Baseline)
    Academic Value: ★☆☆
    Production Quality: ★★★
    Latency: ★★★★★
    Failed Engineers United Club Pvt. Ltd.
    15 / 18
    Vision

    Future Scope

    Eight expansion avenues extending Zentro's intelligence, reach, and capabilities.

    🎙️
    Voice-based AI Assistant
    📱
    Native Mobile Application
    🤝
    Real-time Supplier Negotiation
    🌍
    International Marketplace
    🎯
    Personalized Recommendations
    🔗
    Blockchain Verification
    📈
    AI Demand Forecasting
    🔔
    Price Drop Alerts
    Failed Engineers United Club Pvt. Ltd.
    16 / 18
    Summary

    8-Month Summary Timeline

    Key deliverables at a glance.

    1
    Foundation
    Dev setup, architecture, API research, auth + search
    2–3
    AI Pipeline
    LangGraph, RAG, vector DB, multi-LLM integration
    4–5
    Intelligence
    Comparison engine, recommendations, dashboard
    6–8
    Polish & Launch
    Beta testing, security, deploy, thesis & demo
    32
    Weeks
    4
    Team Members
    ~₹5K
    Total Budget
    100%
    Open Source
    5+
    Marketplaces
    Failed Engineers United Club Pvt. Ltd.
    17 / 18
    Conclusion

    Thank You & Discussion

    Zentro transforms marketplace procurement by using AI to deliver fast, accurate, and intelligent recommendations from a unified platform.

    ⏱️ Saves Time
    💰 Reduces Cost
    ✨ Simplifies Decisions
    📦 Improves Procurement
    🚀 Better UX
    ISBM ISBM College of Engineering, Pune • SPPU SPPU • Department of Artificial Intelligence & Data Science
    ProjectZentro
    DomainAI Marketplace
    UniversitySPPU
    DepartmentAI & Data Science
    OrganizationFailed Engineers United Club
    TeamSnehal, Sahil, Yash, Shubham
    GuideProf. Anil Walke
    FrontendNext.js · React
    BackendFastAPI (Python)
    DatabasePostgreSQL · Qdrant
    AI StackLangGraph · RAG
    DeploymentDocker · Vercel
    Failed Engineers United Club Pvt. Ltd.
    18 / 18
    1 / 18